Jev directory
CommunityJev Models & Reimplementations
Intern-Decision-0.8B
InternLM's 0.8B multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-0.8B: a state, named questions and optional images produce an answer distribution per question in one forward pass. Self-reported seven-suite average 79.38, Brier 0.530.
- Reading time
- 2 min
- Published by
- Community
- Author
- InternLM
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Fine-tuned from Qwen/Qwen3.5-0.8B; each request carries a state, named questions and optional images.
- Answers come from logits read immediately before each <decision> placeholder, not from generate().
- Self-reported seven-suite average 79.38 vs Jev's 88.74; JevBench-Easy 97.92, Hard 52.25, ToolACE 94.52.
- RTX 4090 latency: 33.98 ms mean, 33.44 ms median and 37.50 ms P95 per query, end to end.
- Card promises 1-16 questions per request with up to 62 options each across choice, score and noul types.
Quickstart
python
from inference import DecisionEngine
engine = DecisionEngine(device="cuda")
request = {
"state": "The customer was charged twice and asks for the extra payment back.",
"questions": {
"team": {"type": "choice", "instructions": "Which team should handle this request?",
"criteria": {"billing": "Payments and refunds", "delivery": "Shipping and delivery"}},
"urgency": {"type": "score", "instructions": "Rate the priority.", "criteria": ["Low", "Medium", "High"]},
},
}
response = engine.predict(request)
print(response["answers"])Watch out
Card reviewed 2026-10-17; the benchmark table, Brier/ECE values and RTX 4090 latency figures are the publisher's own measurements, and the card warns latency is workload and hardware dependent. This card reports no calibration pilot, and the Jev 88.74 comparator is the publisher's own run.
Questions about this entry
Short answers drawn from the linked project or page.
InternLM's 0.8B multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-0.8B. It takes a shared state, a schema of named questions and optional images, and returns an answer distribution for every question in one forward pass.
choice (criteria as an ordered object of option values to descriptions), score (a list or an ordered object with finite numeric keys) and noul (binary, options no then yes). The card says to supply 1-16 questions with up to 62 options per question.
Self-reported seven-suite average 79.38 versus Jev's 88.74, with Brier 0.530 and ECE 0.066. JevBench-Easy 97.92, Jevbench-Original 80.56, Jevbench-Hard 52.25, Typed Decision 77.35, ToolACE 94.52, AG News 88.61 and WildJailBreak 64.48.
Per-query end-to-end latency with the local HF inference path: 33.98 ms mean, 33.44 ms median and 37.50 ms P95. The card warns the values are workload and hardware dependent.
All three share the same inference recipe and fine-tune targets at different base sizes. The card's own runs put the seven-suite average at 79.38 here, 84.68 for Intern-Decision-2B and 90.02 for Intern-Decision-4B, so expect stronger decisions as the base grows.
Use Python 3.12+, install requirements.txt in a PyTorch/CUDA environment, then import DecisionEngine and call engine.predict(request); the engine defaults to the checkpoint next to inference.py and returns a Jev-compatible JSON response.
Apache-2.0. The card's benchmark and latency numbers are the publisher's own measurements, and the repository ships the inference module so its default calibration matches the 0.8B checkpoint.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
X: Standard One SH (StandardThinking)
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
InternLM's multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-4B: a shared state, a schema of named questions and optional images go in, and every question returns an answer distribution from one forward pass. The card reports a 90.02 seven-suite average.
Jev Models & Reimplementations#community#open-models#multimodal
Community
InternLM's 2B multimodal structured decision model, fine-tuned from Qwen/Qwen3.5-2B: a state, named questions and optional images produce an answer distribution per question in one forward pass. Self-reported seven-suite average 84.68, Brier 0.437.
Jev Models & Reimplementations#community#open-models#multimodal
Community
A small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
Community
Perplexity's 27B decision model rebuilt on Qwen3.8-27B with a joint request format that answers a whole request's questions in one sequence, reporting 68.09 on Decision Index 0.3 and 77.43 on vision.
Jev Models & Reimplementations#community#open-models#multimodal
Community
A 4B decision model taking text, JSON, or images up to 262,144 tokens: 52.75 on Decision Index 0.2.1, 204 of 231 on JevBench public, and 26.0 ms for a three-question request on one H200.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune from Convai Innovations on the 1,200-case typed-decisions benchmark: it reports 0.777 accuracy on the official 400-case test set, above TypeSafe Jev 1.13.0 at 0.727 and the teacher ceiling at 0.735.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.